Tycho Arete Macro Fund

Overview

Tycho Arete Macro Fund aims to deliver competitive risk-adjusted returns while maintaining a low correlation with all major asset classes. It strives to construct and update macro-analytical frameworks that incorporate the rapidly changing macroeconomic conditions around the world, as well as the significant idiosyncrasies of large global actors such as China and Japan.

Strategy & Manager

Performance

Commentary

Investment Manager’s Commentary – July 2026

In July, the momentum unwind we had flagged in June did realise — but faster, longer, and broader than we had judged likely. What began as a repricing of the global AI hardware complex became, by month-end, a full-scale unwind of the momentum factor itself, worldwide. The scale bears recording: by index measures stretching back to 1999, the drawdown in technology momentum in July was one of the worst on record.

Despite having added hedges and bringing down net exposure across the book, we still suffered large losses in July. That is the signature of a factor event, not a directional one; it is also, we recognise, cold comfort, and we owe investors a precise account of what happened and what we misjudged.

The losses came in two parts. The first was the global AI unwind itself. Our long in global AI hardware bore the brunt as memory and semiconductor names fell violently. The hedges added in June and July — short KOSPI 200, Nikkei and Nasdaq index futures performed as designed, absorbing nearly two-thirds of the gross loss; short KOSPI 200 was the single largest contributor in the book. What we misjudged was not the direction of risk but its duration and velocity: the residual long was sized for a correction of ordinary length, and this one did not stop where prior episodes had. The result was a net loss of roughly 4% on the complex despite the hedge.

The second part came from China — and here the miss was different in kind. Our working assumption, validated as recently as June, was that the bifurcation of the AI ecosystem made China’s onshore complex a diversifier: domestically funded, lightly levered, outside the machinery amplifying the selling elsewhere. July broke that assumption, despite genuinely constructive local developments; Kimi’s frontier progress and CXMT’s capacity ramp, but the onshore market drew no benefit, if anything, cheaper Chinese capability arguably added to the pressure on global hardware economics as it failed to lift the domestic tape.

The unwind was imported, but not as a simple contagion story. Our read is that the onshore market carries its own crowded momentum complex — domestic quantitative strategies concentrated, by construction, in small caps and once the global factor cracked, that positioning unwound in sympathy, through a channel that owed little to AI. Our longs across CSI 1000, STAR 50, ChiNext and China renewables bore the full weight, and our China hedges inverted: the short legs in HK large-cap indices rallied as the onshore longs fell, both sides of the pair losing at once — the same inversion of winners and losers seen globally, and a hedge built against direction offering little protection against factor.

On the other side of the ledger, the cyclical and structural book did its work. Long China financials — insurance foremost — gained as the rotation favoured domestic cash flow over momentum, with China internet and pharma adding alongside. Short US Treasuries contributed as bonds found no flight-to-quality bid — consistent with our long-held view on the secular cost of capital, and a small but telling confirmation that this was a positioning event rather than a growth scare. Gold, our out-of-consensus short, was immaterial on the month.

Current Outlook

What broke the AI trade? It would be tempting to file July under “technical adjustment” and move on — and there would be truth in it since factor drove the month. But an unwind of this degree can imply more fundamental fissures and, not surprisingly, we spent the month focused on the drivers behind the market action and its attendant implications.

Crowding Beyond the Tape

We knew the trade was crowded — hence the hedges added in June, which earned their keep. What we underestimated was that the crowding extended beyond markets, into the real economy itself. At the crux of our misjudgement in July is China’s own leverage to the AI trade. Roughly a third of China’s GDP growth last year came from net exports – a fact that we well understood. What we under-appreciated, however, was that over half of this year’s export growth has come from the AI chain; and within that, first-half chip exports rose 96% by value on just 7% by volume. The growth engine of the world’s second-largest economy, in other words, now runs in meaningful part on memory prices. Add the circular financing of the ecosystem, vendors and customers increasingly funding one another, and a single crack can rapidly lead to wider fissures. A crowded market can be hedged; a crowded economic ecosystem, less so. While the market side has largely cleared (sell-side estimates suggest Korea’s leveraged ETF overhang has shrunk from some $50bn to $17bn, with hedge-fund de-grossing mostly complete), the real-economy imbalances are more appropriately assessed in the context of months and years. The AI trade simply commands a higher risk premium than it did in June, and we now price it accordingly.

The Cycle Question

Memory has a playbook, refined over four decades, and its central rule is sharply etched in the simple math of cyclical supply and demand: the time to sell is when the second derivative turns negative. The peak in the growth rate has always preceded the peak in the price, and the peak in the price has always preceded a painful decline in the equities. By that rule, July delivered the signal. DRAM contract prices rose roughly 90% in the first quarter and 60% in the second; the third-quarter projection is nearer 15% — still rising, but a pronounced flattening of the curve, and one driven not by new supply but by demand beginning to crack.

Yet part of the picture departs from the script: the crack is confined, so far, to the consumer — PC and handset makers – squeezed by months of relentless cost increases. The AI tier shows no such strain — server DRAM remains undersupplied, HBM the most tightly allocated product in the industry, hyperscaler backlogs stretching twelve to eighteen months of contracted demand. This is the debate that will define the next phase of the trade: does AI demand carry the cycle beyond its old playbook, into something structural — a cycle whose marginal buyer is no longer the consumer but the hyperscalers? Possible, and the evidence for it is genuine. But few phrases have cost investors more than “this time is different”, and we choose to respect the cyclical playbook until we see data that proves otherwise. To that end, we remain watchful — above all for any sign of the consumer’s weakness climbing the stack.

When Higher Capex is No Longer Rewarded

What complicates the picture is that even granting firm AI demand — and the hyperscalers’ capital spending still being revised upwards grants it emphatically — the deterioration in free cash flow begets its own question: sustainability. July supplied the test case. Alphabet reported revenue nearly $3bn ahead of expectations, cloud growth of 82%, a backlog beyond half a trillion dollars — and raised its capital spending guidance towards $205bn, tipping free cash flow negative for the first time in its listed history. This would typically lend to the playbook of “Hyperscalers down, hardware up”, but this time is different. For two years the market’s reaction function had been simple — more capex, higher hardware — and in July it snapped: demand confirmed, investment increased, price down. The question had changed, from demand to return, and on the return no one — ourselves included — yet has the answer. The internet scaled at zero marginal cost; intelligence does not. Every token carries a cost of goods, and enterprises have begun to behave accordingly — setting budgets, routing to cheaper models, asking why tenfold the tokens yielded only twice the output. Demand has not fallen; it has ceased to be unconditional. Until the return question finds its answer, capital expenditure reads as cost rather than promise — and this, we judge, is the largest challenge the trade now faces.

Throwing China into the Mix…

July’s other shock came from China: a frontier-grade release from Kimi and a wave of open-weight models good enough to unsettle Wall Street and Washington alike. The instinctive reading — Chinese progress as a direct strike on US AI profitability — mistakes the mechanism. What good-enough open models truly do is commoditise the model layer itself: when a frontier advantage can be distilled away within months, differentiation becomes fleeting, and value migrates elsewhere in the chain — towards compute, and towards whoever owns the user. For hardware demand this is not bearish; commoditised intelligence is consumed in greater volume, the Jevons logic we have argued since the DeepSeek moment. But it hones the return question above to a fine edge: the frontier laboratories must spend ever more to defend leads that erode ever faster. Two ecosystems are consolidating — much as Apple and Android once did — one pursuing the frontier, the other pursuing diffusion, and China’s diffusion path is the cheaper one to walk.

For China itself, however, we hold the uncomfortable thought alongside the constructive one: an economy in which consumption stays soft, growth leans on exports, and exports lean on AI is heading in the right direction while growing more concentrated — and more levered to the very cycle now in doubt. We remain well clear of the China-disruption bears. We also concede that this alone may not be enough to climb the wall of worry.

Where Does This Lead Us

While it is too early to write off the thesis on structurally higher AI demand, markets are increasingly scrutinizing destination vs path. And while our conviction in the demand is intact; our conviction in the path — its smoothness, its multiple — is not, and the positions are now sized to the path rather than the destination. As such, we have shifted our exposure in equities to a more neutral positioning.

On duration and gold, July arguably strengthened the case. Markets may have stopped applauding the capex, but the capex will be spent all the same — contracted, financed, and rising — and capital absorbed on that scale continues to push real yields higher, and Treasuries and gold lower. We retain our conviction on these trades.

Finally, two developments would lead us to reduce further: CSP’s funding stress turning systemic as financial conditions tighten; or the consumer’s weakness climbing the stack into HBM, which would tell us the cyclical reading of memory had defeated the structural one. Neither is present today; both are watched daily.

Documents